On the Use of a Molecular Structure Ontology. Henry Dayringer, Matthias Nolte, Leiming Zhu, and Chris Waller

Transcription

1 On the Use of a Molecular Structure Ontology Henry Dayringer, Matthias Nolte, Leiming Zhu, and Chris Waller

2 User Tier Data Mining and Access (RGate) Client Tier Compound Design (PGVL, MoViT) Project Team Collaboration (SharePoint) enotebook (CeN, Symyx) Mid-tier Bits Server-side Data Integration (PLP, etc.) Service Orchestration (BizTalk, etc.) Data Query Engine Chemical Search Service Computational Services Data and Service Tier 2 CPMart HTSMart SARMart Pfizer File Literature/Patent/ Data Feeds Global Monomers Property Calculation Engine External Data Web Services Model Server

3 Outline What Problem are We Solving? What is an InChI? What is an Ontology? Description of the Molecular Structure Ontology (MSO) Practical Application of the MSO Technology Conclusion?

4 Structure Relationship?s Structure concordance: What sample can I substitute in this assay? Very strict business rules Applies only to internal compounds Related structure: What other compounds are identical (or very nearly so) to compound X? Find related assay data Correlate internal with external compound sources Supports a wide variety of business processes

5 InChI The IUPAC International Chemical Identifier (InChI TM ) is a nonproprietary identifier for chemical substances that can be used in printed and electronic data sources thus enabling easier linking of diverse data compilations. It was developed under IUPAC Project during the period Details of the project and the history of its progress are available from Project Goal: Develop a set of algorithms for the standard representation of chemical structures that will be readily accessible to chemists in all countries at no cost. The standard chemical representation could be used as input into existing and newly developed computer programs to generate a IUPAC name and a unique IUPAC identifier. Source:

6 InChI Description Six InChI Layers Main layer Charge layer Stereochemical layer Isotopic layer Fixed-H layer Reconnected Layer Ambiguous information and specific information can be encoded. Search relevant layers for different levels of concordance

7 Ontology The term ontology has its origin in philosophy, and has been applied in many different ways. The core meaning within computer science is a model for describing the world that consists of a set of types, properties, and relationship types. Exactly what is provided around these varies, but they are the essentials of an ontology. There is also generally an expectation that there be a close resemblance between the real world and the features of the model in an ontology. [3] What ontology has in common in both computer science and in philosophy is the representation of entities, ideas, and events, along with their properties and relations, according to a system of categories. In both fields, one finds considerable work on problems of ontological relativity (e.g., Quine and Kripke in philosophy, Sowa and Guarino in computer science) [4] and debates concerning whether a normative ontology is viable (e.g., debates over foundationalism in philosophy, debates over the Cyc project in AI). Differences between the two are largely matters of focus. Philosophers are less concerned with establishing fixed, controlled vocabularies than are researchers in computer science, while computer scientists are less involved in discussions of first principles (such as debating whether there are such things as fixed essences, or whether entities must be ontologically more primary than processes). Source:

8 Molecular Structure Ontology A derived set of relationships between chemical compounds based on their chemical structures. Specifically: The result of comparisons of InChI string (or parts thereof) representations of chemical structures for the purpose of finding related compounds in a large set of chemical structures.

9 MSO Generation Break MOL file into bonded fragments For each fragment, generate Pipeline Pilot unique tautomer Generate InChI string for each fragment Truncate InChI just before the beginning of stereo information layer Generate an integer hashcode for the full and truncated InChI string (64 bit, collision rate < 1:20,000,000) Register (based on full string and its hash code) unique InChI strings, assign InChI ID Register composition of molecules based on InChI IDs, store repeat counts and assigned type (parent, salt, solvent ) keyed to compound ID

10 MSO Based Lookup Get InChI ID from compound ID or from structure Get all truncated hashcodes that exist with this InChI ID Get all InChI IDs that can produce any of the found truncated hashcodes (expands to all registered full InChI strings that yield the same truncated InChI string) Get all compounds that are known to contain any of these full InChI IDs (with assigned type of parent) Assign the relationship between the input structure and the found structures based on differences in the full InChI strings (different stereo, mirror image, different salt, etc).

11 Related Molecules InChI-based Ontology Partial InChI string Full InChI string Identical structure + stereoisomer code RS PF MN CP RS RS RS PNU UK ST AG PF MN VC ST RS CP MN PF t11-/m0/s1 ST AG PF MN VC t11-/m1/s1 ST RS /{formula}c12h17no2 CP /c{connections}1-9(2)11(13)12(14) MN /h{h_atoms}3-7,9,11h,8,13h2,1-2h3 PF m1/s1/i1+1/t9?,11- /i{isotopic:atoms}1+1 /t{isotopic:stereo:sp3}9?,11- RMCMX: Racemic mixture Chiral SNENK: Single Enantiomer Chiral SNENK: Single Enantiomer /t{stereo:sp3}11- /m{stereo:sp3:inverted}1 /s{stereo:type (1=abs, 2=rel, 3=rac)}1 InChI=1/C12H17NO2/c1-9(2)11(13)12(14) /h3-7,9,11H,8,13H2,1-2H3/t11-/m1/s1/i1+1/t9?,11-/sicSNENK

12 Conceptual: Building a Chemistry Knowledge-Base Knowledge Access Data Mining Knowledge Discovery 1.0 Knowledge 0.8 Presentation P(efficacy) mm µm nm pm log10 Affinity A1 A2 A3 A4 A76 A1 A53 Data Integration Attributes Mapping Cleaning Normalization D1 D2 D135, -0.2, 4.5x10e4, µm, 74.2, Descriptors Reduction D1 D47-2, 0.045, nm, Attributes Descriptors Abstraction

13 Conceptual: Building a Chemistry Knowledge-Base

14 Practical Application of the MSO Tools and opportunities for making the most of the Pfizer structure collection through the MSO technology

15 The Collection The Pfizer structure collection is an aggregate of both internal and external compounds. The internal compounds derive from the aggregation of the compounds of the many legacy companies from which Pfizer is derived. The external compounds derive from a variety of public sources, many of the compounds have associated biological activity data. The Pfizer internal collection has about 11,170,000 distinct parent IDs representing 7,365,000 distinct parent compound structures. The external collection has about 10,104,000 distinct parent IDs representing 7,800,000 distinct parent compound structures. The total collection of 21,275,000 parent IDs representing 14,105,000 distinct compound structures.

16 The Molecular Structure Ontology The service provides the following groupings of compounds related to a starting compound: All compounds that are identical to the starting compound, ignoring stereo chemistry, double bond geometry and isotopic labeling All compounds with the same bonded framework of atoms (with three subgroupings)

17 Framework compounds (experimental) Three categories: 1. Same formula and bonded framework 2. Same framework, different formula 3. Same formula, different framework MSO Relatedness Layers Starting compound MSO compounds Closely related. Same atoms, same bonds. Might differ in stereochemistry, double bond geometry or isotopic labeling

18 MSO Related Compounds Begin with a compound (by ID, etc) Find the compounds very closely related to this compound For example, start with Chantix InChI=1/C13H13N3 /c ( )10(11)4-12(13)15-1 /h1-2,4-5,8-9,14h,3,6-7h2/t8-,9+

19 MSO Related Compound 28 related MSO related compounds including: 3 concorded IDs (business rules apply) 13 identical structures (same salt and stereo, fail some business rule for concordance) 9 that are a different salt 2 with a different chirality 1 labeled compound

20 MSO Related Examples Diff salt Diff stereo Different isotopes Identical parent

21 Current Status: MSO Service Current processes about 400,000 requests per day (mostly SMILES string input) Used to identify existing compounds in database similar to proposed compounds in new synthesis libraries Used to help correlate compounds in research databases (external to internal, external to external) more powerful than canonical SMILES and much faster. Average processing time is about 250 ms per compound.

22 Experimental Don t Try This at Home! New service offering, still in development

23 Framework Related Compounds Framework derived by making all bonds single and all atoms = carbon:

24 Framework Generation Discard all frameworks that reduce to methane (single atom salts, water, etc) From the InChI of the framework, generate a hashcode Register the full InChI string and the hashcode for unique frameworks Register each unique molecular formula Register the composition of each molecule correlated to its framework and formula 20,000,000 compounds yielded about 7,000,000 unique frameworks

25 Framework Results Maximum frequency is 465,000 occurrences for ethane Next most frequent is 71,000 occurrences for n-butane About 1,100 (of about 7,000,000) frameworks occur 500 times or more

26 Framework Searching From the input ID (or structure), get the framework ID assigned and the molecular formula of the original structure component Find all other compounds in the database with the same framework ID or the same formula Return three sets: Same framework and formula Same framework, different formula Same formula, different framework

27 Same Framework, Same Formula There are 13 compounds in database with the same framework and formula (after removing those related by the MSO) as the input molecule: Chantix 6 have EXT IDs, 7 have internal IDs. Some examples (similarity scores calculated using the Accelrys SDK): Red highlights the areas of change between reference and comparison compound Similarity=.57

28 Same Framework, Different Formula (23 compounds identified, some examples show below) Similarity=.23 Similarity=.42 Similarity=.63 Similarity=.88

29 Same Formula, Different Framework 766 compounds identified, some examples show below. Highlighted part comes from a lenient maximum common substructure analysis. Similarity=.11 Similarity=.19 Similarity=.29 Similarity=.31

30 Preliminary Conclusions The MSO service provides a unique capability to locate compounds related to a starting compound of interest. The service is very fast The capability is based on the use of the novel properties of the InChI structure representation

31 User Tier Data Mining and Access (RGate) Client Tier Compound Design (PGVL, MoViT) Project Team Collaboration (SharePoint) enotebook (CeN, Symyx) Mid-tier Bits Server-side Data Integration (PLP, etc.) Service Orchestration (BizTalk, etc.) Data Query Engine Chemical Search Service Computational Services Data and Service Tier 31 CPMart HTSMart SARMart Pfizer File Literature/Patent/ Data Feeds Global Monomers Property Calculation Engine External Data Web Services Model Server

32

33 Review: What Have We Done With This? Simplified complex structure-based queries (any atom, any bond, S-groups, etc.) Simplified complex business rules from an information technology perspective (no Cheshire, no structure checker scripting, no complex stereochemistry codes, etc.) Simplified complex framework analysis methodology

34 Future: What Might We Do With This? Simplify complex Oracle structure cartridge technology ( grep ) Discontinuous substructure (pharmacophore) search issues Simplify the complex patent searching/filing/ defending process To be determined

35 Thanks Henry Dayringer, Matthias Nolte, Leiming Zhu, and Chris Waller